The Junior Lawyer Problem and the Promise Built on It
Legal teams have long operated under a quiet structural lie: that every contract gets read. The OpenAI customer story on Legora opens with the honest version of that fiction — a junior lawyer handed an impossible stack of agreements overnight, able to focus on only a fraction of them. The rest pass through without scrutiny, and everyone involved understands this without saying it aloud. Legora’s founding insight, as described in the customer story, is that legal teams are unusually time crunched, which means some valuable work simply does not get done. That is not a technology problem. It is a capacity problem dressed up as a diligence problem.
The promise that follows is seductive precisely because it addresses the unspoken gap. If AI can review the contracts nobody had time to read, then the fiction becomes true. The platform Legora builds — what CEO and co-founder Max Junestrand calls the “agentic operating system for legal work” — is presented as the mechanism by which every piece of legal work in the world could eventually run on one system The ambition is enormous, and it rests on a claim that deserves scrutiny: that the machine’s first pass is a real pass, not a plausible-looking substitute for one.
The Numbers That Measure Speed, Not Judgment
The headline figures in the Legora story are striking and precise. Legora’s own legal team cut the share of its capacity spent reviewing NDAs from roughly 20 % to under 1 %. A Fortune 100 customer reduced contract turnaround time by 90 %, saved more than $20 million on external litigation and diligence, and returned 9.5 hours per lawyer each week for higher-value work. These are the numbers the company chooses to publish, and they are the numbers a reader will remember.
Look closely at what they actually measure. Turnaround time measures how long a document takes to move through a process. Hours returned measures how much labor was displaced. Money saved on external counsel measures what was not spent. None of these figures measures whether the review caught the clause that mattered. The gap between what the metrics show and what a legal team actually needs to know is the central tension in the entire Legora story, and it is a tension the customer story never directly confronts. Speed is easy to instrument. Judgment is not.
Where the Machine’s First Pass Comes From
The NDA workflow is described in concrete terms, and the concreteness is instructive. General Counsel Emma Kadri Bergström’s team analyzed the agreements they had historically signed and built a playbook that lets Legora take a first pass at incoming NDAs, flagging terms that fall outside acceptable parameters. This is a rule-following system trained on past decisions. It works because the past decisions were consistent enough to be encoded.
The honesty of that description is worth pausing on. The system does not understand an NDA. It compares one against a template derived from precedent. When the incoming document resembles the precedent, the flagging is reliable. When it does not — when a counterparty introduces a novel structure, an unusual indemnity, a clause that means something different in a new context — the system’s confidence is not calibrated to its competence. It will still produce a first pass. The question is whether the lawyer receiving that pass knows which parts of it to distrust.
The Five-Minute Fix and the Illusion of Diagnosis
CTO Jacob Lauritzen describes how coding agents changed the engineering team’s ability to respond when something goes wrong. Previously, diagnosing a product issue could mean several engineers investigating different possible causes over a number of hours. Now, coding agents can search millions of lines of logs and pursue many possible explanations simultaneously, sometimes taking Legora from bug report to working fix in around five minutes.
That is a genuine capability shift, and it is also a case study in the difference between generating an answer and generating the right answer. A system that pursues many possible explanations at once will generate many candidate fixes. The speed of generating them is not the same as the reliability of choosing among them. The customer story presents the five-minute fix as a straightforward gain. A more careful reading suggests a new kind of risk: a fix applied before anyone has fully understood the cause, and a team that learns to trust the speed because the speed is what it can see.
What the Customer Never Sees
The customer story is candid about one thing: customers do not see the engineering efficiency itself. What they experience is faster resolution, earlier feedback, and ideas tested sooner. This is presented as a virtue — the customer gets the benefit without the machinery. But it also describes a structural opacity. The customer’s experience of the product is shaped by processes the customer cannot inspect, and the trust required to accept that is considerable.
That opacity compounds when the product itself is the thing making judgments. Alex Fortescue-Webb, Global Head of Legal Engineering, points to due diligence as an example, and Legora’s Tabular Review workflow can analyze thousands of contracts across multiple questions in minutes, allowing lawyers to apply detailed scrutiny much more broadly. The stated benefit is that risks previously hidden in the manually unreviewed subset can now surface. The unstated risk is the inverse: a lawyer who trusts the broad sweep may stop looking at the contracts the sweep did not flag. The tool expands coverage and simultaneously creates a new blind spot, and the blind spot may be invisible in practice.
The Quality Claim and Its Missing Evidence
One line in the customer story carries more weight than any of the metrics: “AI doesn’t just make you faster. It also increases quality This is the claim that would justify everything else. If the machine’s first pass is not merely faster but better, then the capacity problem is genuinely solved rather than displaced.
The evidence offered for that claim is thin. The due diligence example shows that broader review can surface risks outside the manually reviewed subset, which is a coverage argument, not a quality argument. The Fortune 100 figures are time and money figures. Nowhere in the customer story is there a measurement of error rates, false negatives, or the proportion of flagged items that a senior lawyer would have judged material. The quality claim is asserted, it is plausible, and it remains an assertion. That is the gap in its purest form: a claim about judgment supported by evidence about throughput.
The Rebuilt
Roadmap and the Moving Target Lauritzen’s description of his own workflow is a revealing passage in the story. Model improvements force him to reconsider the roadmap on a biweekly basis. Engineers, product managers, designers and sales teams all frequently ask him what the new thing is that has become possible for customers.
Read that again as a statement about reliability. The product’s capabilities are being redefined faster than any customer can build stable expectations around them. A legal team that designs a workflow around what the system does in March may find the system doing something different in June, not because the workflow was wrong but because the underlying intelligence moved. This is presented as momentum. It is also a description of a system whose behavior is not fixed, and a legal practice built on a tool whose behavior is not fixed may be resting on unstable ground.
The Bottleneck That Was Never Only About Speed
The framing of the stakes is a human moment in the source. Every legal department is worried about becoming a bottleneck, the story notes, and that is when the business stops coming to you.
The logic is appealing and it contains a hidden assumption: that the reason legal teams were avoided was slowness, and that removing slowness removes the avoidance. But legal teams are also avoided because they say no, because they raise risks, because they slow a deal down for reasons that turn out to be correct. Arguably, a legal function that becomes fast and agreeable is not necessarily a legal function that is doing its job. The transformation from bottleneck to partner may be real, and it may also be a transformation from friction to consent, and the two are not the same thing.
The Overnight
Diligence and the Morning Review Lauritzen’s picture of the near future is vivid: agents tackling longer-running tasks that require far greater context and autonomy, due diligence spanning millions of pages running overnight, with the analysis ready for a lawyer to review the next morning. Alongside that, increasingly capable models becoming faster, enabling real-time collaboration where lawyer and AI work back and forth on a contract rather than treating AI as a task-and-answer tool.
Both futures are plausible, and both sharpen the central problem rather than resolving it. An analysis of millions of pages delivered overnight is only as useful as the lawyer’s ability to verify it, and verification of a machine’s work at that scale is itself a task of enormous size. The more autonomous the agent, the more the lawyer’s morning review becomes an act of trust rather than an act of checking. The system’s output will look like diligence. Whether it is diligence depends on scrutiny the system’s own scale makes harder to perform.
The Perception Problem and the Real One
The closing argument in the source is about perception. If Legora succeeds, the outcome will not simply be lawyers completing today’s workload more efficiently but legal teams able to take on work that capacity once made impossible, with more time for judgment, strategy and the high-value decisions that still require a human.
That last clause is doing a great deal of work. It concedes that some decisions require a human, and it does not say how the boundary between machine work and human work gets drawn, who draws it, or how anyone would know when the boundary has been crossed. Junestrand’s closing line — that it is the best time in history to be a lawyer — is a promise about the profession’s future. The honest version of that promise would come with a question attached: best for which lawyers, doing which work, and judged by whom.
The Question the Metrics Cannot Answer
The
Legora story is a story about a real capability meeting a real need, and the numbers it reports are company-reported figures about time and money. None of that is in dispute. What is in dispute is whether the numbers answer the question that matters. A legal system’s value is not the speed at which it processes documents or the hours it returns to a calendar. It is whether the advice that comes out the other end is right, and whether the people relying on it can tell when it is not.